| Summary: | 碩士 === 國立中興大學 === 農業機械工程學系 === 90 === The image processing techniques can be applied for the classification of variety and quality of agricultural products. The classification parameters include length, width, perimeter and area. The recognition function of the inspection will be improved depending on the precision and raised due to measured data. In this study, CCD camera with four scales 1/64, 1/16, 1/4 and 1 of resolutions was utilized to acquire images of rice kernels. The noise of image was removed by the median filter. The moment-preserving principle was applied to determine the thresholds automatically to reduce the effect of changes of the acquiring environment on length, width, perimeter and area.
A micrometer (10mm×10mm, 0.1mm grid) was applicated to calculate the length and width of rice using the projection image of each rice kernel. Two methods, the automatic thresholding and the sobel operator with LoG method were employed to detect the edge of rice image and to calculate its length and width. The experimental results obtained from both methods were then compared with that using the fixed threshold method.
The results show that the automatic thresholding is able to reduce the effect of variation of the light source. The use of LoG method can improve the resolution and measurement precision from pixel resolution to subpixel resolution. According to the Scheffe test of SAS, the percentage of errors measured by the automatic thresholding with LoG method were improved at scales of 1/64 (13.79×10-2 mm/pixel) and 1 (1.69×10-2 mm/pixel). To increase the inspection amount of rice kernels per image and to improve the inspection speed without reducing the precision, the automatic thresholding with LoG method is a superior way of edge detection for the four scales considered in this research. The choose of resolution can be determined by the requirement of the inspection precision, for instance, 108 rice kernels per image at scale 1/64 (13.79×10-2 mm/pixel) and 30 rice kernels per image at scale 1/16 (6.79×10-2 mm/pixel) that can achieve the inspection accuracy required.
|
|---|
